A NOVEL PPDM PROTOCOL FOR DISTRIBUTED PEER TO PEER INFORMATION SOURCES

Document Sample
A NOVEL PPDM PROTOCOL FOR DISTRIBUTED PEER TO PEER INFORMATION SOURCES Powered By Docstoc
					International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
 INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING &
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME
                                TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)                                                     IJCET
Volume 4, Issue 4, July-August (2013), pp. 358-380
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)                  ©IAEME
www.jifactor.com




      A NOVEL PPDM PROTOCOL FOR DISTRIBUTED PEER TO PEER
                     INFORMATION SOURCES

                  S Kumaraswamy1, Manjula S H1, K R Venugopal1, L M Patnaik2
                        1,2
                          Department of Computer Science and Engineering
              1
               University Visvesvaraya College of Engineering, Bangalore University,
                                       Bangalore 560 001
                   2
                     Honorary Professor, Indian Institute of Science, Bangalore.


ABSTRACT

        Cryptographic approaches are traditional and preferred methodologies used to preserve the
privacy of data released for analysis. Privacy Preserving Data Mining (PPDM) is a new trend to
derive knowledge when the data is available with multiple parties involved. The PPDM deployments
that currently exist involve cryptographic key exchange and key computation achieved through a
trusted server or a third party. The key computation over heads, key compromise in presence of
dishonest parties and shared data integrity are the key challenges that exist. This research work
discusses the provisioning of data privacy using commutative RSA algorithms eliminating the
overheads of secure key distribution, storage and key update mechanisms generally used to secure
the data to be used for analysis. Decision Tree algorithms are used for analysis of the data provided
by the various parties involved. We have considered the C5.0 data mining algorithm for analysis due
to its efficiency over the currently prevalent algorithms like C4.5 and ID3. In this paper the major
emphasis is to provide a platform for secure communication, preserving privacy of the vertically
partitioned data available with the parties involved in the semi-honest trust model. The proposed Key
Distribution-Less Privacy Preserving Data Mining (                  ) model is compared with other
protocols like Secure Lock and Access Control Polynomial to prove its efficiency in terms of the
computational overheads observed in preserving privacy. The experiential evaluations proves the
              reduces the computational overheads by about 95.96% when compared to the Secure
Lock model and is similar to the computational overheads observed for the Access Control
Polynomial model.

Keywords: Privacy Preserving Data Mining, Semi Honest Model, Secure Multiparty Computation,
Commutative RSA, C5.0 Data mining Algorithm, Classification Rules, Key Distribution.



                                                358
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

1. INTRODUCTION

         Varied parties like research organizations, government agencies, and business houses
maintain data for analysis. The knowledge derived from their local data repositories is insufficient to
meet their projected outcomes. Hence there exists a need for sharing data for effective data mining
and better analysis. Privacy of the data available across varied parties released for analysis is of
primary importance in a PPDM System. For example to curb terrorism there exists a need to analyze
the data available with the immigration department of each country to understand the movement and
track terrorists and suppress acts of terrorism. With the help of such cooperative systems in place,
nations use the data to track and analyze the threats that terrorism poses to humanity and security of
its citizens. Additionally such mechanisms can be utilized to monitor illegal immigrants. The data
analysis is used for constructive or economy enhancing activities like tourism and business.
Generally it is observed that the parties involved are not comfortable to disclose the entire
information involved in spite of agreements and strategies in place to preserve privacy of the data.
There are number of organizations like the United States Health Insurance Portability and
Accountability in the United States of America and European Union Privacy Directive in Europe
which have prescribed the norms to be followed prior to data released for analysis to preserve the
privacy of the data [1][2][3][4][5]. To address these issues this paper proposed a novel PPDM
mechanism namely                 .The local data available with the parties involved in
model is not released or shared for analysis. Instead the classification rules generated at the local
parties is used for analysis thus preserving privacy of the data available with each party. The data
available at each party is assumed to be vertically partitioned and all the parties involved in the
proposed mechanism are semi honest in nature.
   Decision Tree classification algorithms have been frequently used by researchers for accurate
analysis. The data mining accuracy of the proposed system depends on the rules locally generated.
The                 utilizes the C5.0 Data Mining Algorithm for accurate rule generation and
classification. The use of cryptography is adopted in the                to provide privacy to the data
released for analysis. In the               proposed in this paper the locally generated are secured
using cryptographic techniques namely commutative RSA. The use of commutative RSA is adopted
to overcome the drawbacks of key distribution, rekeying overheads and key compromise threats.

1.1. Motivation
       The parties involved in the PPDM model governed by the semi honest trust model are not
comfortable with the aspect of data release for analysis. To preserve the privacy the data researchers
have adopted cryptographic techniques relying on the keys for encryption and decryptions. The keys
are symmetric or asymmetric in nature. In order to derive the keys and distribute them amongst the
parties the use of a third party or a secure server is considered [6][7][8]. The use of a sure third party
server induces overheads related to key computation, key storage and key distribution is known as
the protocol initialization phase. Post the initialization phase the data is secured using cryptography
and is shared for analysis. The design of                 draws its motivation targeted to eliminate the
need of a third party server, minimise the network operation overheads (i.e. key storage and key
distribution) and eliminate the need for the actual local data available with the parties to be released
for analysis.

1.2. Contribution
      The research work introduces the             model. The              considers the local rules
generated using the C5.0 data mining algorithm to be released by each party for analysis and not the
actual data. The privacy of the locally generated rules is preserved using the commutative RSA
cryptographic algorithm. The              minimizes the computational overheads associated with

                                                   359
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

every privacy preserving model (namely key distribution, key storage and key computation using
external entities). The privacy of the locally generated rules is preserved and is proved using the
computationally indistinguishablity [9][10][11] property. The              model proposed preserves
privacy in the presence of dishonest or non trusted parties.

1.3. Organization
       The remaining paper is organised as follows. The background and related work is discussed in
the second and the third sections of the paper. Section four of the paper introduces the Vertically
Partitioning of data. The commutative RSA algorithm is explained in fifth section of this paper. The
proposed               model is discussed at length in the next section. The comparisons of the
proposed               with Secure Lock and Access Control Polynomial protocol in terms of the
computational complexity is discussed in section seven of the paper. The eighth section proves the
computationally indistinguishablity of the                thus preserving privacy. The conclusion and
future of the research work is discussed in the last section of this paper.

2. RELATED WORK – PRIVACY PRESERVING DATA MINING

       Research highlighting the benefits of data mining was introduced in the early 1980’s and was
adopted by business originations and other establishments a decade later. Provisioning of privacy of
the data to be utilized for mining or PPDM architectures found its emergence in the beginning of the
20th century [12] [13]. There afterwards researchers proposed numerous PPDM models to secure
data and facilitate data mining put forth the limitations that exist in the PPDM systems. Nan Zhang
and Wei Zhao [14] discuss that PPDM systems are broadly classified based on the privacy level the
systems provide. The classifications are namely Secure Multi Party Computation Techniques and
Partial Information Hiding techniques.
       Y. Lindell and B. Pinkas [15] proposed a PPDM system in which the parties are bound by the
semi honest trust model [10]. The privacy of the data is maintained even in the presence of colluded
users. The PPDM model utilized the ID3 classification mining algorithm. The major drawback of
this system is that the model is limited to two parties and the classification accuracy of the ID3
algorithm is lower compared to the C4.5 and C5.0 data mining algorithm.
       In secure multiparty computation techniques proposed by researchers is based on the varied
data distribution techniques adopted amongst the parties involved. The data available with the parties
is horizontally partitioned as considered by C. Clifton et al [13], R. Agrawal and R. Srikant [16],
Yaping Li et al., [17], Ming-Jun Xiao et al., [18]. The major drawback of these systems is that the
models described cannot be extended to data which is generally vertically partioned. The data mining
algorithms adopted exhibit lesser classification accuracy when compared to the C5.0 data mining
algorithm.
       O. Goldreich [19], A.W.-C. Fu et al., [20], J. Vaidya and C. W. Clifton [21] considered
vertically portioning of the data available with the parties involved. The proposed models provided
for data integrity of the information available with the various parties involved. The major drawback
of these systems is the reduced data mining accuracy of these systems.
       The partial information hiding can be further classified into three categories namely data
perturbation, k-anonymity and retention replacement. D. Agrawal and C. C. Aggarwal [22][23] , K.
Chen and L. Liu [24] and S. Papadimitriou et al., [25] adopted the data perturbation technique to
preserve the privacy of the data to be released for mining. The major drawback of these approaches
is that the data perturbation techniques adopted effect the data mining results.
         To overcome the drawbacks of data perturbation privacy preserving technique L. Sweeney et
al., [26], C. C. Aggarwal and P. S. Yu [27], Slava Kisilevich et al.,[28] proposed the use of k-


                                                 360
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

anonymity privacy preserving technique. The use of these techniques exhibit a high degree of data
annomization and reduced mining accuracy.
         W. Du and Z. Zhan[29], R. Agrawal, R. Srikant, and D. Thomas[30] in their papers
introduced retention replacement techniques to preserve privacy. The retention replacement
techniques elements with a certain probability are retained as it is or the elements are replaced based
on the type of elements and probability distribution function adopted for that type of element. The
major drawback of these systems is that such privacy preserving techniques can be adopted only for
data that is continuous in nature. In general partial information hiding technique is not capable of
effectively performing in the presence of colluded users hence the approach has not been adopted in
the proposed approach.
        Knowledge extraction or data mining is a vital operational requirement of any successful
PPDM system. The                    protocol presented in this paper adopts C5.0 decision tree algorithm
to the purpose of mining [31][32][33][34]. The use of C5.0 classification tree algorithm is preferred
over its predecessors decision tree algorithms like ID3 used by J. Vaidya and C. Clifton [21],
Xindong Wu et al., [32] and C4.5 adopted by Ming-Jun Xiao et al.,[18], Yanguang Shen et al., [35]
due to the mining accuracy it provides[32].
         In addition to the above mentioned techniques there have been efforts to provide data privacy
utilizing introduction of noise to preserve privacy Po-Hsun Sung et al., [36], C. Dwork et al.,
[37].Ineffective de-noising techniques and inability to obtain original data is the major drawback of
this approach. The data available with the parties have been secured by using specific anonomization
techniques introduced by Matthews et al., [5]. The annonomization technique adopted by Matthews
et al., [5] effect the mining results hence such annomizations techniques have not been adopted.
         The use of cryptographic techniques to secure data transmitted for mining by Yaping Li [17] ,
Ming-Jun Xiao et al., [18] , R. Agrawal and R. Srikant [16] , B. Pinkas [38] have found to be very
effective and do not effect the mining accuracy as desired in PPDM systems. The major drawback of
these systems is the use of traditional data mining algorithms which render limited mining results.
         The novelty of the proposed PPDM protocol is evident as it not only provides security using
commutative algorithms [39][11] and eliminates the overheads arising due to key distribution , key
storage overheads and re-keying[40][41][42].

3. BACKGROUND WORK

         G. H. Chiou et al., [6] proposed the “Secure Lock” secure group communication protocol.
This protocol considers an external third party server for key computation and key distributation. The
secure lock protocol is based on the asymmetric cryptographic protocol and utilizes a lock based
mechanism to secure the data transactions over the network. The major drawback of the secure lock
protocol is that the protocol is computationally heavy and considers a third party server for the
initialization phase. To overcome the drawbacks of the computationally cumbersome Secure Lock
protocol Xukai Zou et al., [7] [8] proposed the Access Control Polynomial protocol. The Access
Control Polynomial introduces a novel key management scheme adapted to support multi party
group communications. The Access Control Polynomial Protocol is computationally lighter than the
Secure Lock Mechanism as it adopts the symmetric cryptographic techniques to preserve privacy and
data integrity. The Access Control Polynomial considers an external central server for the protocol
initialization which contributes to key computation, key storage, key management and key exchange
overheads there by reducing its efficiency.
         J. Vaidya et al., [43][21] and R. Agrawal et al., [16] considered the ID3 decision tree data
mining algorithm for analysis of the data released by the parties in the PPDM model. The drawback
ID3 data mining algorithm is that it cannot handle missing parameters in datasets, continuous data,
provides no support for pruning. To overcome the drawbacks researchers Ming-Jun Xiao et al., [18]

                                                  361
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

and J. Vaidya et al., [21][43] utilized the C4.5 algorithms to attain higher classification accuracy and
handle continuous data. The C4.5 algorithm was improved by J. Ross Quinlan [31][32]33][34] to
obtain higher classification accuracy and reduce the execution time.

4. VERTICALLY PARTITIONED DATA

      This paper highlights a PPDM protocol when the data available with the parties involved is
vertically partitioned. Vertically partitioned data or heterogeneous distribution of data occurs when
the parties involved possess data of independent attributes and all the attributes available with the
parties composite the complete transactions to be utilized for mining. Let us consider a set of
parties involved in a semi honest trust model represented by the set
Let the data to be utilized for Mining      can be represented as



Where Represents the total number of transactions and
      Represents the each transaction of the mining data
Each Transaction       is represented as



Let the data available with the   party be represented as



Where                      Represents the attributes available with Party
The   Data is said to be vertically partitioned if

                                                 }

        For example let us analyze the effects of pollution on heart diseases. Data is available with
the Pollution control Board and the health ministry. The pollution control board provides an area
wise pollution parameters and the health departments provides the data related to the personal in that
area having heart related ailments. With the help of these two data it is possible to study the effect of
pollution on heart ailments.

5. COMMUTATIVE RSA

       Provision of privacy of the data released for mining is an important functionality to be
provided in any PPDM system. Cryptography is the preferred means to provide for privacy of the
data. In this paper we have adopted the commutative property of the RSA Algorithm for provisioning
of privacy. We shall now discuss the construction proof and the commutative nature of the RSA
algorithm adopted in the            . The proposed protocol assumes that the parties involved pre
exchange two prime numbers        and prior for key generation amongst themselves as a prerequisite
for the construction of the PPDM system considered. The prime numbers selected are randomly
decided such that        .
       Let us consider 2 parties named Aruna and Jahnavi to have exchanged two prime numbers
and such that           .The RSA is an asymmetric cryptographic algorithm so Aruna and Jahnavi


                                                  362
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

generate the encryption and decryption keys based on the following equations. The encryption key is
represented as       and the decryption key as      .




It’s observed that the obtained by Aruna and Jahnavi is similar, public and does not reveal the pre
computed prime numbers and . Now to obtain each party selects a large random number such
that it is a co prime of . I.e. the following equation must be satisfied to find the co prime.



The parameter     is computed using the following equation.



Let us consider a data . The encryption of the data using the RSA Algorithm is represented as
                   And the decryption operation is represented as



Where    represents the decrypted data.
        Let us consider that the encryption keys of Aruna and Jahnavi are                      and the
decryption keys are represented as                 . To prove the commutative nature of the algorithm
we need to prove that for the considered data represented as , if Aruna encrypts the data first and
then Jahnavi encrypts the data the resultant is equivalent to resultant obtained when Jahnavi encrypts
the data and then Aruna encrypts the data again. Let represent the commutative RSA operation
and the data considered to be represented by .




6. KEY DISTRIBUTION-LESS PRIVACY PRESERVING DATA MINING (KDLPPDM)
   SYSTEM

       This section of the paper discusses the proposed             . The protocol adopts a three step
approach apart from the system initialization step. In the system initialization step each party
generates mining rules locally using the C5.0 algorithm on the pre classified data available with each
party. The initialization step additionally discusses commutative RSA algorithm initialization
wherein each party computes their encryption and decryption keys.




                                                 363
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

The List of symbols used in the paper shown in the Table1 provided below.

                           Table 1. Notations used in the Algorithms
        SYMBOLS          DEFINITION
                         A Set of parties involved in the PPDM.
                         The th party of the set         .
                         Set of parties excluding       or initiator party.
                         Vertically partitioned data of   parties.
                         Pre classified data.
                         Classification rules obtained from the data dt.
                         Rules Classification set.
                         Rule Set.
                         Classification rules obtained from C5.0 algorithm.
                         Combined classification rules.
                         Computation overhead in PPDM protocol initialization.
                         Computation overhead in computing ,phi,GCD.
                         Computation overhead in computing Encryption and Decryption keys.
                         Party initiator used to construct combined rule pool.
                         Secure rule pool.
                         Maximum number of rules obtained from party
                         Represent the communication function
                         Classification algorithm function of the C5.0 data mining algorithm.
                         Combined rule file generation function.
                         C5.0 Classification function.
                         Total data bits transacted.
                         Unclassified data available with
                         Communication cost of step 2
                         Classification rules for n parties
                         Classification rules
                         The max number of rules obtained from party           .
                         Rules count.
                         Decryption key of each party.
                         Encryption key of each party.
              ,          Two prime numbers and where                   .
                         Set of number of transctions.
                             th
                                number transaction or total number of transactions in set   .
                          th
                         t unique attribute.

The steps of the               are as follows
   i. The provisioning of privacy of the locally generated rules and construction of the secure
      combined rule pool. The secure combined rule pool is a collection of the locally generated rules
      by each party in an encrypted form to preserve privacy.
  ii. Obtaining the combined rule pool to use these rules for mining and analysis. In this step one
      party is initialized as the initiator who propagates the secure rule pool amongst the various
                                                  364
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

       parties. The parties involved decrypt the secure rule pool using the commutative RSA
       decryption key. The initiator who decrypts the secure rule set finally obtains the combined rule
       set.
  iii. The analysis and the mining result of the unclassified data using the C5.0 classification
       algorithm.
         We now discuss the preliminary notations used to describe the model, the initialization of the
PPDM protocol and the three step approach adopted in the realization of the                           is
discussed.

6.1. Preliminary Notations used in KDLPPDM
       Let us consider the set            represents the     parties involved in the PPDM system
considered. It is assumed that the      parties involved agree to participate in a Semi-Honest Trust
Model. The set        is defined as follows



The set       can be defined as



Where          and
 Let            represent a set of parties in the PPDM system excluding party           , defined by




  Let the set  represent the vertically partitioned data available with all the      parties involved in
the PPDM system defined by



  The proposed PPDM protocol assumes that each party contains pre classified data and unclassified
data. The data       consists of      transactions which contain     attributes. The data available with
the parties is vertically partitioned .i.e. each of the parties have      transactions set available with
them but no two parties have the same attribute appearing in their dataset. The data available with the
     party is defined as



And              Where represents the total number of transactions
Each transaction     contains unique attributes represented by



The data available with the       party represented as



                                                  365
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

        Where              . As                   and                 .We can state that
and                          as the data available with each party of the set              is vertically
partitioned. The data set       represents the pre classified data set       available with each party
                 . Each party         contains a data set        which represents an unclassified data
which needs to be classified. As the locally generated rules provide lesser mining accuracy [44] the
parties of the set      agree to share the locally generated rules to achieve a higher degree of mining
accuracy in an semi-honest trust model[13][17][38].

6.2. KDLPPDM System Initialization
        The PPDM system discussed in this paper considers that the data          is vertically partitioned
and the pre classified data           available with the                       is used to generate the
classification rules using the C5.0 data mining algorithm. The C5.0 algorithm is preferred owing to
its higher classification accuracy when compared to the commonly used classification tree algorithms
like ID3[21][32] and C4.5[13][12][43]. Let                    represent the C5.0 classification function
and                and       represents the classification rules obtained from the data . The rule
generation function based on the pre classified data      and the classification set

                        is defined as

Where     the rule set constructed is based on the data      and the classification set   .

Algorithm :                 Initialization

Input: Two prime numbers        and     where
Output: Encryption key =                        , Decryption key =                    Local classification
rules,   for each party

   1. Require 2 prime numbers and where
   2. For each party
   3.     Initialization of          Privacy Preserving Function
   4.
   5.
   6.        Compute
   7.         Compute
   8.         Obtain
   9.        Obtain
   10.       Party          encryption key =
   11.      Party          decryption key =
   12.     End initialization of           Privacy Preserving Function
   13.    Initialization of                 C5.0 Rule Generation
   14.         Obtain Pre classified data set of
   15.         Obtain Classification Set
   16.         Compute local classification rules
   17.   End Rule Generation
   18. End for each.


                                                    366
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

        The rule set     generated locally are released by all the parties of the set       for analysis.
Preserving the privacy of the locally generated rule set is achieved by adopting the advantages of
Commutative RSA. There is a need to generate the encryption and decryption keys for each party. To
obtain the keys using commutative RSA the parties exchange two prime numbers and where
        .Let           represent the privacy preserving function incorporated in the proposed PPDM
system , using commutative RSA. The greatest common divisor function is denoted as                      .
The system initialization phase is described by the algorithm given above.
        On completion of the initialization each party         of the set       possesses its encryption
and decryption keys for preserve the privacy of the rules          generated using the C5.0 algorithm.
The locally generated rules are utilized in the construction of the secure rule pool using commutative
RSA as discussed in the next section of this paper. The objective of the proposed PPDM protocol is
be stated as the construction of the combined rule pool          defined by preserving privacy



Where represents party                    . The       is utilized to obtain the mining results on the
unclassified data set     .
  The locally generated rules        are provided by each party for analysis. The actual data
available with party is not compromised and not shared amongst the varied parties involved even
though there exists a cryptographic protocol (Commutative RSA) in place to provide privacy like in
the prevalent systems [45][46] in place.
  The computation overheads represented involved in the PPDM protocol initialization is defined as
follows


  Where      represents the computational overhead observed in terms of the computational times
involved and represents the computational overhead applicable in computing        , and the      .
    represents the computational overhead in computing the encryption and decryption keys. Let
      represent the time complexity function involved in solving bit operations. The computational
overhead is defined as follows



6.3. Step 1: Privacy Provisioning of the Classification Rules and Construction of the Secure
   Rule Pool
        The foremost step of the proposed PPDM protocol is targeted towards the construction of the
secure rule pool represented by          and the privacy preserving feature of the locally generated
rules. To provide for privacy each party releases the locally generated rules       after encrypting
using its encryption key                  . All the other parties encrypt the rules       using their
respective encryption keys. The commutative encryption function for data       using the encryption
key        is defined as


Extending the above definition to a multi part scenario the encryption function performed by party
defined as

Let           represent the secure rule pool to be constructed. The secure rule pool defined as


                                                  367
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

Where represents the total number of parties involved in the PPDM model and                 represents the
secured rule set of    obtained after number of encryptions.

The privacy provisioning of the locally generated rule set is achieved using the algorithm mentioned
below

Algorithm Name : Secure Rule Pool Construction

Input: Encryption key =                      ,Local classification rules    of each party
Output: Secure Rule Pool

   1. Initialize
   2. For each
   3.     Compute
   4.     Compute
      where
   5.     For Each
   6.            Compute
   7.     End For
   8.
   9. End For each


         From the above algorithm it is evident that the rules available with each party are encrypted
times. To construct the secure rule pool there exists an communication cost involved owing to the
transfers of the secure rules amongst the parties involved. The communication cost                     of
the first step defined as


Where         represents communication function represents the parties involved and represents
the total data in bits transacted.
        This is a very critical step involved as this step provides the privacy of the data using
commutative cryptography approaches. In the case of a malicious party involved the system is
designed such that the even on obtaining any data, the obtained data is in an cryptic format with zero
knowledge as to how many times the data is encrypted there by providing privacy. More over the
data transacted amongst the parties is computationally indistinguishable. For better understanding
let’s consider parties denoted by                and                         and             represent
their encryption and decryption functions respectively. Let the data available with                 be
represented as . The data          available with              is to be provided to      .For privacy
provisioning the parties            provide the data to by encrypting the data with their respective
encryption functions represented by                      . Privacy is provided if
received by       are computationally indistinguishable. i.e.                    . The computational
indistinguishablity of the proposed PPDM protocol is discussed in the future sections of the paper.

6.4. Step 2: Construction of the Combined Rule Pool maintaining Privacy
      This step of the proposed PPDM protocol is targeted towards the construction of the
combined rule set from the secure rule set constructed using step 1


                                                  368
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

Decryption of the secure rule pool at each party is achieved using the Commutative RSA algorithm
utilizing their respective decryption keys         . The commutative decryption function of the
encrypted data is defined as


Extending the above definition to a multi party communication scenario for a party      defined as



Let      represent the C5.0 classification rules of the party                . The combined rule pool
is defined as


From the combine rule pool algorithm it is clear that the initiator           propagates the secure rule
pool          amongst all the parties          bound in the considered                model. Each party
decrypts the secure rules of the secure rule pool and then sends it to the next party in such a way that
the initiator decrypts the secure rule pool last. The initiator receives the secure rule pool decrypted
exactly            times. The final decryption of the secure rule pool performed by the initiator
provides the combined rule pool         .

The algorithm adopted in the construction of the combine rule pool is as mentioned below.

Algorithm Name : Combined Rule Pool Construction

Input: Decryption key =                     of each party and Secure Rule Pool
Output: Combined Rule Pool

   1. Initialize
   2. Initialize initiator
   3. Compute
       where
   4. For each
   5.      Initialize
   6.          For each secure rule
   7.              Compute
   8.
   9.          End For
   10.
   11. End For
   12. Initialize
   13. For each secure rule
   14.      Compute
   15.
   16. End For
   17.



                                                  369
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

The communication cost                of the second step is defined as



Where         represents communication function represents the parties involved and represents
the total data in bits transacted. The bits transferred            .
  Though this step any party is capable of construction the combined rule pool              securely
without worrying about privacy as the transacted data is always in a cryptic form providing security
even in the presence of malicious parties. Moreover all the parties involved are unaware of the
encryption and decryption keys utilized in the                 model as there is no key exchange
involved. The computational indistinguishablity of this step is proved in the results and discussion
section of this paper.

6.5     Step 3: Analysis using the C5.0 Algorithm using the Combined Rule Pool
        The first two steps of the              model concentrated on provisioning of the privacy of
the classification rules exchange. This step discusses the data mining algorithm adopted for analysis
of the unclassified data available with the party. Once the rules generated by all the parties are
accumulated, these rules are combined to generate stronger classification rules utilized in mining.
    Let us consider          represents the unclassified data available with                  . Let the
combined rule file generation utilizing all the rules of the combined         be defined as




From the above definition it is clear that the combined rule file      embodies all the rules of
           The max number of rules obtained from party              is represented as            Where
        represents the combined rule file generation function.
Let the classification algorithm function of the C5.0 data mining algorithm be represented as



Where           represents the C5.0 classification function. The classification function considers the
classification rules       and the unclassified data        available with party as inputs. The output
of the function is denoted as                   .
The classification data                     represented in the form of decision trees is utilized for
analysis of the unclassified data
        The               model discussed in this paper preserves the privacy of the data released for
analysis and provides a rule based classification analysis platform using the advanced C5.0 decision
tree algorithm [31][32][33][34]. Data available with each party involved is kept secure and no party
releases that data for analysis instead only the rules generated are exchanged for analysis.



                                                 370
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

The data mining process of the                model is achieved on the basis of the algorithm given
below

Algorithm Name : Data Analysis using the C5.0 Mining Algorithm

Input: Combined Rule Pool
Output: Data mining analysis results

   1.    Initialize
   2.    Initialize
   3.    For each
   4.         For each
   5.
   6.
   7.        End For
   8.
   9.    End For
   10.   Compute

        The proposed model eliminates the communication overheads arising from key exchange and
at the same time preserves privacy as no party involved in the PPDM model exchanges any key
amongst themselves owing to the adoption of Commutative RSA algorithm in our model. In the case
of malicious parties involved in the system there is no notable data risk, as the data transacted over
the network is in the cryptic form and is computationally indistinguishable proved in the subsequent
sections of this paper. The next section of the paper discusses the computational complexity of the
             model constructed using a tri step approach.

7. COMPUTATIONAL ANALYSIS AND COMPARISONS

       The amount of resources utilized in solving a computational algorithm or problem desired is
known as the computational analysis. In this section of the paper the computational analysis of our
proposed                model is discussed. Furthermore this section discusses the computational
efficiency of the              model over the existing secure group communication models namely
Secure Lock and Access Control Polynomial. The computational efficiency is measured in terms of
the time complexity involved in solving the provided protocol on a homogenous computing
environment. For comparisons the initialization step of the protocols is considered. In Secure Lock
[6] and the Access Control Polynomial [7][8] a central server is considered for computing the
cryptographic keys utilized for secure communication amongst the parties considered. The central
server computes the keys and distributes them securely to the parties involved in communication.
The                model does not consider a external or third party secure server to overcome the
drawback of key distribution and key compromise attacks.
       The PPDM initialization step discussed in the section six of this paper is responsible for the
key computation and key derivation of the                   model. The computational complexity
involved depend on the number of parties involved and the data transacted for key establishment and
initialization. To prove that the            model proposed through this paper is computationally
less expensive when compared to the Secure Lock and Access Control Polynomial model we shall
consider number of parties involved in communication and each identified by an identity
represented as        . The proposed Secure Lock communication protocol is based on the public and
private key cryptography. Access Control Polynomial is based on symmetric key cryptography.
                                                 371
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

      The                model utilizes the benefits of Commutative RSA cryptography protocol for
secure communication. Let us consider              represents the private public key pairs applicable to
the secure lock protocol. L             represents the symmetric key applicable to the Access Control
Polynomial protocol and                represent the encryption decryption key pairs applicable to the
             model. It is evident that greater the key size greater is the computation cost involved. A
1024 bit asymmetric cryptography is similar to an 80 bit symmetric key cryptography scheme [47]
hence we consider the key lengths of                and             of 1024 bits in length and the key
lengths of           of 80 bit to use for comparative analysis. Let the computation function be
represented by          . Let’s consider the Initialization, Key Computation and Derivation, Group
Membership Verification, Encryption/Decryption Cost and storage cost at the server for
comparisons.

The computational costs observed are as shown in the Table2 provided below.

                      Table 2. Algorithm based Computational Cost Analysis
        PHASE OF               SECURE LOCK           ACCESS
      ALGORITHM                                     CONTROL
                                                  POLYNOMIAL
       Initialization

  Key Computation and
       Derivation

   Group Membership
      Verification

 Encryption/Decryption
   Server End Storage


         All the above discussed models have been developed on the Visual Studio 2010 platform.
The implementations were carried out using C#.Net. From Table 1 it is clear that the
model does not consider the Group Membership Verification Phase and the Server End Storage
computational overhead as there is no server involved. For comparisons the number of parties
involved were varied form 10,20,30,50 70 and 100. The computational complexity of the
Initialization Phase and the Key Computation and Derivation Phase of all the 3 models were
measured in terms of the time involved in computation. The implementations were tested on an Intel
Core 2 Duo 2.40 GHz CPU having 4GB of RAM. The Initialization, Key Computation and
Derivation and the Group Membership Verification phases of the 3 algorithms have been considered.
The results obtained are represented graphically in Figure 1 shown below. The
Encryption/Decryption phase and the Server End Storage phase have been neglected for the analysis
presented here.
         The results obtained are as shown in Figure 1 it is evident that the Secure Lock protocol is
computationally more expensive than the Access Control Polynomial and the secure group
communication algorithm proposed in this paper are computationally less expensive than the Secure
Lock Scheme of secure group communication. The secure group communication protocol proposed
in this paper provides more security and is less vulnerable to data loss as it does not consider any key
exchange amongst the data custodians. The                  described in this paper does not consider a

                                                  372
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

central trusted server for group establishment and cryptographic key distribution. In case of any
malicious data custodians the data exchanged using Secure Lock and Access Control Polynomial is
more vulnerable as the data transacted during communication is encrypted only once but in our
proposed scheme the probability of the data transacted is encrypted multiple times providing for
more secure means of communication.




                    1400
                                COMPUTATIONAL TIME ANALYSIS
                                   SECURE LOCK
                    1200
                                   ACCESS CONTROL POLYNOMIAL
                    1000           KDLPPDM
       Time in ms




                    800

                    600

                    400

                    200

                       0
                           10         20             30           50              70   100
                                           Number of Data Custodians Considered

                                  Figure 1. Computational Time Analysis

8. COMPUTATIONALLY INDISTINGUISHABLE ANALYSIS AND COMMUTATIVE
   NATURE PROOF OF KDLPPDM

        The proof of privacy preserving provisioning is provided based on the computationally
indistinguishablity [9][10][11] analysis of the data distribution observed in the proposed PPDM
protocol. To demonstrate the proof of computational indistinguishablity a prototype PPDM system
based on the              was developed using the Visual Studio 2010 platform on the Microsoft .Net
framework 4.0. The prototype implementation was considered assuming that the number of parties
involved in the semi honest trust model is 3. The data set considered for analysis was the
hyperthyroid data set. The hyperthyroid data set was vertically partitioned amongst the 3 parties
considered. The C5.0 algorithm was utilized to generate the local rules at each parties end based on
the vertical partitioned data each party housed. The C5.0 algorithm was run on an Linux Platform
[31]. The commutative RSA algorithm was initialized in accordance to the                 Initialization
algorithm discussed in the sixth section of the paper. On completion of the protocol initialization the
locally generated rules were encrypted at each parties end and the creation of the secure rule pool
was considered using Step 1 discussed earlier. The data distribution graphs of the transmissions post
the secure rule pool creation is considered to provide the computational indistinguishablity proof.
The data distribution frequency versus the data size graphs obtained are as shown




                                                      373
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME


                             140                   Data Distributation of Orignal Data
                             120
  Distributation Frequency




                             100
                              80
                              60
                              40
                              20
                              0
                                         1
                                        26
                                        51
                                        76
                                       101
                                       126
                                       151
                                       176
                                       201
                                       226
                                       251
                                       276
                                       301
                                       326
                                       351
                                       376
                                       401
                                       426
                                       451
                                       476
                                       501
                                       526
                                       551
                                       576
                                       601
                                       626
                                       651
                                       676
                                       701
                                       726
                                       751
                                       776
                                                                   Data Size (Kb)
  Figure 1. Data Distribution of Locally Generated Data Mining Rules for Data Custodian

                             7000        Data Distributation of Data Encrypted n Times (n = 1)
                             6000
  Distributation Frequency




                             5000
                             4000
                             3000
                             2000
                             1000
                                   0
                                         1
                                        26
                                        51
                                        76
                                       101
                                       126
                                       151
                                       176
                                       201
                                       226
                                       251
                                       276
                                       301
                                       326
                                       351
                                       376
                                       401
                                       426
                                       451
                                       476
                                       501
                                       526
                                       551
                                       576
                                       601
                                       626
                                       651
                                       676
                                       701
                                       726
                                       751
                                       776
                                                                    Data Size (Kb)
                               Figure 2. Data Distribution of Data Custodian a ’s rules Encrypted n times using
                                                         Commutative RSA

                             8000        Data Distributation of Data Encrypted n Times (n = 2)
                             7000
  Distributation Frequency




                             6000
                             5000
                             4000
                             3000
                             2000
                             1000
                                   0
                                         1
                                        26
                                        51
                                        76
                                       101
                                       126
                                       151
                                       176
                                       201
                                       226
                                       251
                                       276
                                       301
                                       326
                                       351
                                       376
                                       401
                                       426
                                       451
                                       476
                                       501
                                       526
                                       551
                                       576
                                       601
                                       626
                                       651
                                       676
                                       701
                                       726
                                       751
                                       776




                                                                    Data Size (Kb)
Figure 3. Data Distribution of Data Custodian                              rules Encrypted   times using Commutative
                                        RSA


                                                                     374
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME


                             7000   Data Distributation of Data Encrypted n Times (n = 3)
                             6000
  Distributation Frequency




                             5000
                             4000
                             3000
                             2000
                             1000
                                0
                                      1
                                     26
                                     51
                                     76
                                    101
                                    126
                                    151
                                    176
                                    201
                                    226
                                    251
                                    276
                                    301
                                    326
                                    351
                                    376
                                    401
                                    426
                                    451
                                    476
                                    501
                                    526
                                    551
                                    576
                                    601
                                    626
                                    651
                                    676
                                    701
                                    726
                                    751
                                    776
                                                            Data Size (Kb)
Figure 4. Data Distribution of Data Custodian ’s rules Encrypted                 times using Commutative
                                        RSA

                             8000   Data Distributation of Data Decrypted n Times (n = 1)
                             7000
  Distributation Frequency




                             6000
                             5000
                             4000
                             3000
                             2000
                             1000
                                0
                                      1
                                     26
                                     51
                                     76
                                    101
                                    126
                                    151
                                    176
                                    201
                                    226
                                    251
                                    276
                                    301
                                    326
                                    351
                                    376
                                    401
                                    426
                                    451
                                    476
                                    501
                                    526
                                    551
                                    576
                                    601
                                    626
                                    651
                                    676
                                    701
                                    726
                                    751
                                    776
                                                            Data Size (Kb)
Figure 5. Data Distribution of Data Custodian ’s rules Decrypted                 times using Commutative
                                        RSA

                             7000   Data Distributation of Data Decrypted n Times (n = 2)
                             6000
  Distributation Frequency




                             5000
                             4000
                             3000
                             2000
                             1000
                                0
                                      1
                                     26
                                     51
                                     76
                                    101
                                    126
                                    151
                                    176
                                    201
                                    226
                                    251
                                    276
                                    301
                                    326
                                    351
                                    376
                                    401
                                    426
                                    451
                                    476
                                    501
                                    526
                                    551
                                    576
                                    601
                                    626
                                    651
                                    676
                                    701
                                    726
                                    751
                                    776




                                                           Data Size (Kb)
Figure 6. Data Distribution of Data Custodian P’s rules Decrypted                times using Commutative
                                        RSA

                                                             375
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME



                    140           Data Distributation of Data Decrypted n Times (n = 3)
                    120
  Distributation Frequency




                    100

                             80

                             60

                             40

                             20

                              0
                                    1
                                   25
                                   49
                                   73
                                   97
                                  121
                                  145
                                  169
                                  193
                                  217
                                  241
                                  265
                                  289
                                  313
                                  337
                                  361
                                  385
                                  409
                                  433
                                  457
                                  481
                                  505
                                  529
                                  553
                                  577
                                  601
                                  625
                                  649
                                  673
                                  697
                                  721
                                  745
                                  769
                                  793
                                                         Data Size (Kb)

Figure 7. Data Distribution of Data Custodian ’s rules Decrypted            times using Commutative RSA


      Based on the above Figures it is clear that the proposed algorithm is computationally
indistinguishable and it could be observed that Figure 2 is identical to Figure 8 , Figure 3 is identical
to Figure 7 and Figure 4 is identical to Figure 6 which proves the commutative nature of the
proposed system.

9. CONCLUSIONS AND FUTURE WORK

        PPDM is utilized to derive knowledge when the data is housed in a distributed environment.
This paper introduces the Key Distribution-Less Privacy Preserving Data Mining (
model to preserve the privacy and provides an environment to attain desired knowledge extraction
when the data available is vertically partitioned and distributed amongst the various parties involved.
It is assumed that all the parties involved unite under a semi-honest trust model to achieve their
respective mining goals. The                  introduced adopts the commutative RSA cryptographic
algorithm to secure the data and preserve its privacy. The commutative property of the RSA
algorithm eliminates the overheads and the risks involved in key computation, key distribution, key
storage and key exchange even in the presence of colluded parties. The parties are reluctant to share
the original local data amongst one and other, hence the proposed                beleives in sharing the
locally generated data mining rules providing for enhanced privacy preserving features. The use of
the C5.0 data mining algorithm is adopted in the                   to achieve higher mining accuracy,
higher speed and for rule generation. The experimental evaluation presented proves that the proposed
             model and the Access Control Polynomial reduce the computational complexity by
about 95.6% when compared to the Secure Lock mechanism. The privacy preserving feature of the
             is proved by the computational indistinguishablity analysis provided. The future of the
research work presented in this paper is targeted towards an improved in depth understanding the
C5.0 algorithm and to prove its efficiency over the existing data mining algorithms.




                                                           376
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

REFERENCES

  [1]    N.R. Adam and J.C. Wortmann, (1989) “Security-Control Methods for Statistical
         Databases: A Comparison Study”, ACM Computing Surveys, Vol. 21, No. 4, pp. 515-556.
  [2]    D.G. Marks, (1996) “Inference in MLS Database,” IEEE Transactions on Knowledge and
         Data Engineering, Vol. 8, No. 1, pp. 46-55.
  [3]    M.-S. Chen, J. Han, and P. S. Yu, (1996) “Data mining: An overview from database
         perspective,” IEEE Transactions on Knowledge and. Data Engineering, Vol. 8, No. 6, pp.
         866–883.
  [4]    C. Clifton and D. Marks, (1996) “Security and Privacy Implications of Data Mining”, Proc.
         1996 ACM SIGMOD Int'l Workshop Data Mining and Knowledge Discovery, pp. 15-16.
  [5]    Matthews, Gregory J., Harel, Ofer, (2011) “Data confidentiality: A review of methods for
         statistical disclosure limitation and methods for assessing privacy”, Statistics Surveys, Vol.
         5, pp. 1-29.
  [6]    G. H. Chiou and W. Chen, (1989) “Secure broadcasting using the Secure Lock”, IEEE
         Transactions on Software Engineering, Vol. 15, No. 8, pp. 929–934.
  [7]    X. Zou, Y.-S. Dai, and E. Bertino, (2008) “A practical and flexible key management
         mechanism for trusted collaborative computing”, Proceedings of IEEE INFOCOM'08,
         Phoenix, AZ, USA, pp. 1211–1219.
  [8]    Xukai Zou, Mingrui Qi, and Yan Sui,( 2011) “A New Scheme for Anonymous Secure
         Group Communication”, System Sciences (HICSS) 44th Hawaii International Conference.
  [9]    Oded Goldreich, (2001) “Foundations of Cryptography: Basic Tools”, Vol. 1. Cambridge
         University Press.
  [10]   Oded Goldreich, (2004) “Foundations of Cryptography: Basic Applications”, Vol. 2.
         Cambridge University Press.
  [11]   Alexandre Evfimievski and Tyrone Grandison, (2009) "Privacy-Preserving Data Mining",
         Encyclopedia of Database Technologies and Applications, IGI Global, pp. 1-8.
  [12]   V.S. Verykios et al., (2004) “State-of-the-Art in Privacy Preserving Data Mining”,
         SIGMOD Record, Vol. 3, No. 1, pp. 50-57.
  [13]   C. Clifton et al., (2003) “Tools for Privacy Preserving Distributed Data mining”, SIGKDD
         Explorations, Vol. 4, No. 2, pp. 28-34.
  [14]   Nan Zhang and Wei Zhao, (2007) "Privacy-Preserving Data Mining Systems" ,IEEE
         Computer, Vol. 40, No. 4, pp. 52-58.
  [15]   Y. Lindell and B. Pinkas, (2000) “Privacy preserving data mining”, Springer-Verlag
         Advances in Cryptology, pp. 36-53.
  [16]   R. Agrawal and R. Srikant, (2000) “Privacy preserving data mining” In Proceedings of the
         ACM SIGMOD Conference on Management of Data, pp. 439-450.
  [17]   Yaping Li, Minghua Chen, Qiwei Li and Wei Zhang, (2012) “Enabling Multilevel Trust in
         Privacy Preserving Data Mining”, IEEE Transactions on Knowledge and Data
         Engineering, Vol. 24, No. 9, pp. 1598–1612.
  [18]   Ming-Jun Xiao, Kai Han, Liu-Sheng Huang and Jing-Yuan Li, (2006) "Privacy Preserving
         C4.5 Algorithm Over Horizontally Partitioned Data", Proceedings of the Fifth IEEE
         International Conference on Grid and Cooperative Computing, pp. 78-85.
  [19]   O. Goldreich, (2002) “Secure multi-party computation,” Final (incomplete) draft, version
         1.4.
  [20]   A.W.-C. Fu, R. C.-W.Wong, and K.Wang, (2005) “Privacy-preserving frequent pattern
         mining across private databases”, Fifth IEEE International Conference on Data Mining
         (ICDM'05), pp. 613-616.


                                                 377
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

  [21] J. Vaidya and C. W. Clifton, (2002) “Privacy preserving association rule mining in
        vertically partitioned data,” Proceedings of the eighth ACM SIGKDD international
        conference on Knowledge discovery and data mining (KDD’02), pp. 639-644.
  [22] D. Agrawal and C. C. Aggarwal, (2001) “On the design and quantification of privacy
        preserving data mining algorithms”, Proceedings of the twentieth ACM SIGMOD-SIGACT-
        SIGART symposium on Principles of database systems (PODS '01), pp. 247-255.
  [23] A. Evfimievski, R. Srikant, R. Agrawal, and J. Gehrke, (2002) “Privacy preserving mining
        of association rules”, Proceedings of the eighth ACM SIGKDD international conference on
        Knowledge discovery and data mining(KDD '02), pp. 217-228.
  [24] K. Chen and L. Liu, (2005) “Privacy preserving data classification with rotation
        perturbation”, Fifth IEEE International Conference on Data Mining.
  [25] S. Papadimitriou, F. Li, G. Kollios, and P. S. Yu, (2007) “Time series compressibility and
        privacy”, Proceedings of the 33rd VLDB International Conference on Very Large
        Databases (VLDB '07), Vienna, Austria.
  [26] L.Sweeney, (2002) “k-anonymity: A model for protecting privacy”, International Journal
        of Uncertainty, Fuzziness and Knowledge-Based Systems (IJUFKS), Vol. 10, No. 5, pp.
        557-570.
  [27] C. C. Aggarwal and P. S. Yu, (2004) “A condensation approach to privacy preserving data
        mining,” Proceedings of the International Conference on Extending Database Technology
        (EDBT), vol. 2992, pp. 183-199.
  [28] Slava Kisilevich, Lior Rokach, Yuval Elovici and Bracha Shapira, (2010) "Efficient
        Multidimensional Suppression for K-Anonymity", IEEE Transactions on Knowledge and
        Data Engineering, Vol. 22, No. 3, pp. 334-347.
  [29] W. Du and Z. Zhan, (2003) “Using randomized response techniques for privacy-preserving
        data mining”, Proceedings of the Ninth ACM SIGKDD International Conference on
        Knowledge Discovery and Data Mining, pp. 505-510.
  [30] R. Agrawal, R. Srikant, and D. Thomas, (2005) “Privacy preserving OLAP”, Proceedings
        of the. ACM SIGMOD International Conference. On Management of Data, pp. 251-262.
  [31] http://www.rulequest.com/see5-info.html
  [32] Xindong Wu, Vipin Kumar, J. Ross Quinlan, Joydeep Ghosh, Qiang Yang, Hiroshi
        Motoda, Geoffrey J. McLachlan, Angus Ng, Bing Liu and Philip S. Yu, et al. (2008) "Top
        10 algorithms in data mining", Springer, Knowledge and Information Systems Vol. 14, No.
        1, pp. 1-37.
   [33] Tomasz Bujlow, Tahir Riaz, Jens Myrup Pedersen, (2012) "A method for classification of
        network traffic based on C5.0 Machine Learning Algorithm", in Workshop on Computing,
        Networking and Communications, IEEE, pp 237-241.
  [34] Meng Wang, Kun Gao , Li-jing Wang and Xiang-hu Miu, (2012) “A Novel Hyperspectral
        Classification Method Based on C5.0 Decision Tree of Multiple Combined Classifiers”,
        Fourth International Conference on Computational and Information Sciences (ICCIS), pp.
        373- 376.
  [35] Yanguang Shen, Hui Shao and Li Yang, (2009) "Privacy Preserving C4.5 Algorithm over
        Vertically Distributed Datasets", IEEE, International Conference on Networks Security,
        Wireless Communications and Trusted Computing, Vol.2, pp. 446-448.
  [36] Po-Hsun. Sung, Jyh-Dong Lin, Shih-Huang Chen, Shun-Hsing Chen, and Jr-Hung Peng,
        (2010) “Utilization of Data Mining on Asset Management of Freeway Flexible Pavement”,
        Proceedings of the IEEE International Conference on Industrial Engineering and
        Engineering Management (IEEM), Vol.10, pp. 977 – 979.



                                              378
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

  [37] C. Dwork, K. Kenthapadi, F. McSherry, I. Mironov, and M. Naor, (2006) “Our data,
       ourselves: Privacy via distributed noise generation”, Advances in Cryptology-
       EUROCRYPT, pp. 486–503.
  [38] B. Pinkas, (2002)" Cryptographic techniques for privacy-preserving data mining", ACM
       SIGKDD Explorations Newsletter, Vol. 4, No. 2, pp. 12-19.
  [39] Chris Clifton, Murat Kantarcioglu, Jaideep Vaidya, Xiaodong Lin, and Michael Y. Zhu
       “Tools for Privacy Preserving Distributed Data Mining”,SIGKDD Explorations,Vol. 4, No.
       2, pp 1-7.
  [40] Victor P. Hubenko Jr., Richard A. Raines, Rusty O. Baldwin, Barry E. Mullins, Robert F.
       Mills, and Michael R. Grimaila,(2007) “Improving Satellite Multicast Security Scalability
       by Reducing Rekeying Requirements”,IEEE Network ,Vol. 21, No.4, pp. 51-56.
  [41] Bezawada Bruhadeshwar and Sandeep S. Kulkarni, (2011) "Balancing Revocation and
       Storage Trade-Offs in Secure Group Communication", IEEE Transactions on Dependable
       and Secure Computing, Vol. 8, No. 1, pp. 58-73.
  [42] Nathaniel Karst, and Stephen B. Wicker, (2011) "On the Rekeying Load in Group Key
       Distributions Using Cover-Free Families ", IEEE Transactions on Knowledge and Data
       Engineering, Vol. 58, No. 10, pp. 6667 – 6671.
  [43] J. Vaidya and C. Clifton, (2003) “Privacy-preserving k-means clustering over vertically
       partitioned data,” Proceedings of the ninth ACM SIGKDD international conference on
       Knowledge discovery and data (KDD '03), pp. 206-215.
  [44] Patrick Sharkey, Hongwei Tian, Weining Zhang, and Shouhuai Xu, (2008) “Privacy-
       Preserving Data Mining through Knowledge Model Sharing”, Springer-Verlag ,Berlin
       Heidelberg 2008 LNCS 4890, pp. 97–115.
  [45] Gabriel Ghinita, Panos Kalnis, and Yufei Tao, (2011) “Anonymous Publication of Sensitive
       Transactional Data”, IEEE Transactions on Knowledge and Data Engineering, Vol. 23, No.
       2,. Pp. 161-174.
  [46] Pui K. Fong and Jens H. Weber-Jahnke, (2012) “Privacy Preserving Decision Tree
       Learning Using Unrealized Data Sets”, IEEE Transactions on Knowledge and Data
       Engineering, Vol. 24, No. 2, pp. 353 – 364.
  [47] Arjen K. Lenstra, “Key length. Handbook of Information Security”, Editor-in-Chief,
       Hossein Bidgoli, pp. 617–635.
  [48] Khatri Nishant P., Preeti Gupta and Tusal Patel, “Privacy Preserving Clustering on
       Centralized Data Through Scaling Transformation”, International Journal of Computer
       Engineering & Technology (IJCET), Volume 4, Issue 3, 2013, pp. 449 - 454, ISSN Print:
       0976 – 6367, ISSN Online: 0976 – 6375.
  [49] D.Pratiba and Dr.G.Shobha, “Privacy-Preserving Public Auditing for Data Storage Security
       in Cloud Computing”, International Journal of Computer Engineering & Technology
       (IJCET), Volume 4, Issue 3, 2013, pp. 441 - 448, ISSN Print: 0976 – 6367, ISSN Online:
       0976 – 6375.
  [50] Sumana M and Hareesha K S, “Preprocessing and Secure Computations for Privacy
       Preservation Data Mining”, International Journal of Computer Engineering & Technology
       (IJCET), Volume 4, Issue 4, 2013, pp. 203 - 212, ISSN Print: 0976 – 6367, ISSN Online:
       0976 – 6375.
  [51] R. Manickam, D. Boominath and V. Bhuvaneswari,, “An Analysis of Data Mining: Past,
       Present and Future”, International Journal of Computer Engineering & Technology
       (IJCET), Volume 3, Issue 1, 2012, pp. 1 - 9, ISSN Print: 0976 – 6367, ISSN Online:
       0976 – 6375.



                                             379
                                      Engineering
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976      0976-
                                                         July August
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 4, July-August (2013), © IAEME

BIOGRAPHIES

                   Kumaraswamy S is currently working as an Assistant Professor in the
                                t
                   Department of Computer Science and Engineering, KNS Institute of Technology,
                   Bangalore, India. He obtained his Bachelor of Engineering from SiddaGanga
                   Institute of Technology, Bangalore University, Tumkur. He received his M E
                                                             Engineering
                   Degree in Computer Science and Engineering from UVCE, Bangalore
                   University, Bangalore. He is presently pursuing his Ph.D programme in the area
                        rivacy                                             University.
                   of privacy management in databases in Bangalore University His research
                   interest is in the area of Data mining, Web mining and Semantic web.


                                 ula
                    S H Manjula is currently the Chairman, Department of Computer Science
                    and Engineering, University Visvesvaraya College of Engineering, Bangalore
                    University, Bangalore. She obtained her Bachelor of Engineering and Masters
                    Degree in Computer Science and Engineering from University Visvesvaraya
                                            (UVCE).
                    College of Engineering (UVCE). She was awarded Ph.D in Computer Science
                    from Dr. MGR University, Chennai. Her research interests are in the field of
                    Wireless Sensor Networks and Data mining.


                    K R Venugopal is currently the Principal, University Visvesvaraya College
                                                    rincipal,
                     of Engineering, Bangalore University, Bangalore. He obtained his Bachelor of
                     Engineering from University Visvesvaraya College of Engineering. He received
                     his Masters degree in Computer Science and Automation from I     Indian Institute
                     of Science Bangalore. He was awarded Ph.D in Economics from Bangalore
                     University and Ph.D in Computer Science from Indian Institute of Technology,
                                      distinguished
                     Madras. He has a distinguished academic career and has degrees in Electronics,
                                  Law,
                     Economics, Law, Business Finance, Public Relations, Communications,
Industrial Relations, Computer Science and Journalism. He has authored 39 books on Computer
Science and Economics, which include Petrodollar and the World Economy, C Aptitude, Mastering
                  r
C, Microprocessor Programming, Mastering C++ and Digital Circuits and Systems etc.. During his
three decades of service at UVCE he has over 350 research papers to his credit. His research interests
                                                                       Distributed
include Computer Networks, Wireless Sensor Networks, Parallel and Distributed Systems, Digital
Signal Processing and Data Mining.

                   L M Patnaik is currently Honorary Professor, Indian Institute of Science,
                                                        Chancellor
                   Bangalore, India. He was a Vice Chancellor, Defense Institute of Advanced
                   Technology, Pune, India and was a Professor since 1986 with the Department of
                   Computer Science and Automation, Indian Institute of Science, Bangalore.
                   During the past 35 years of his service at the Institute he has over 500 research
                                                                                  Proceedings.
                   publications in refereed International Journals and Conference Proceedin He is
                   a Fellow of all the four leading Science and Engineering Academies in India;
                   Fellow of the IEEE and the Academy of Science for the Developing World. He
                                                                                          Technica
has received twenty national and international awards; notable among them is the IEEE Technical
Achievement Award for his significant contributions to High Performance Computing and Soft
Computing. His areas of research interest have been Parallel and Distributed Computing, Mobile
                                                                         Neuroscience.
Computing, CAD for VLSI circuits, Soft Computing and Computational Neuroscience.

                                                 380

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:9/5/2013
language:
pages:23